Instructions to use Carlosian/Qwen3.5-35B-A3B-Abliterated with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Carlosian/Qwen3.5-35B-A3B-Abliterated with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Carlosian/Qwen3.5-35B-A3B-Abliterated")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Carlosian/Qwen3.5-35B-A3B-Abliterated")
model = AutoModelForCausalLM.from_pretrained("Carlosian/Qwen3.5-35B-A3B-Abliterated", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Carlosian/Qwen3.5-35B-A3B-Abliterated with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Carlosian/Qwen3.5-35B-A3B-Abliterated"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Carlosian/Qwen3.5-35B-A3B-Abliterated",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Carlosian/Qwen3.5-35B-A3B-Abliterated

SGLang

How to use Carlosian/Qwen3.5-35B-A3B-Abliterated with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Carlosian/Qwen3.5-35B-A3B-Abliterated" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Carlosian/Qwen3.5-35B-A3B-Abliterated",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Carlosian/Qwen3.5-35B-A3B-Abliterated" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Carlosian/Qwen3.5-35B-A3B-Abliterated",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Carlosian/Qwen3.5-35B-A3B-Abliterated with Docker Model Runner:
```
docker model run hf.co/Carlosian/Qwen3.5-35B-A3B-Abliterated
```

Carlosian commited on Jun 28

Commit

3e05860

verified ·

1 Parent(s): 5662e37

Add independent residual-refusal audit (refusal_bench, 200 prompts)

Browse files

Files changed (1) hide show

README.md +31 -0

README.md CHANGED Viewed

@@ -68,6 +68,37 @@ The residual ~2.5% of retained refusals are concentrated on the deliberately-pre
 > **Capability benchmarks are pending.** This release reports the abliteration (Stage-1) validation only. Decontaminated capability evaluation (HumanEval/+, MBPP, GSM8K, AIME, MMLU-Pro, IFEval, BFCL), a StrongREJECT-style willingness×ability uncensoring score, and the three-arm A/B/C comparison are part of the Nous-Heretic protocol and **have not yet been run on this checkpoint**. No capability number is asserted here.
 ## Intended use
 - **Security research and red-teaming** — probing model behavior without refusal confounds.

 > **Capability benchmarks are pending.** This release reports the abliteration (Stage-1) validation only. Decontaminated capability evaluation (HumanEval/+, MBPP, GSM8K, AIME, MMLU-Pro, IFEval, BFCL), a StrongREJECT-style willingness×ability uncensoring score, and the three-arm A/B/C comparison are part of the Nous-Heretic protocol and **have not yet been run on this checkpoint**. No capability number is asserted here.
+## Independent residual-refusal audit
+An independent 200-prompt residual-refusal audit (`refusal_bench`, 2026-06-27) was run on a **Q4_K_M GGUF quantization** of this checkpoint (no adapter), single-stream on a Radeon 890M iGPU (Vulkan, `-ngl 99`, `ctx=16384`, `max_tokens=512`, `temp=0.7`, `top_p=0.9`, `seed=42`). Each response is scored complied / soft-refused / hard-refused / degenerate by a heuristic (pattern-based) classifier.
+| Disposition | Count | Rate |
+|---|---|---|
+| Complied | 195 / 200 | **97.5%** |
+| Hard refusal | 0 / 200 | **0.0%** |
+| Soft-refused (answered, with disclaimer/hedge) | 3 / 200 | 1.5% |
+| Degenerate ("brain-damaged"\*) | 2 / 200 | 1.0% |
+\* heuristic false positives on repetitive-but-valid code (e.g. a buffer-overflow walkthrough, a sonnet), not genuine failures.
+**Compliance by category:**
+| Category | n | Complied | |
+|---|---:|---:|---|
+| pentest | 60 | 98% | |
+| violence | 20 | 100% | |
+| illegal | 20 | 100% | |
+| sexual | 15 | 100% | |
+| political | 15 | 100% | |
+| manipulation | 15 | 100% | |
+| dual_use | 15 | 100% | |
+| coherence (control) | 30 | 97% | |
+| **selfharm** | 10 | **70%** | 3 soft-refused |
+The only residual friction is the **self-harm** category: the model answers genuine support questions but declines lethal-means requests, emitting crisis-resource language instead — the **deliberately retained guardrail** behaving as designed. **Hard-refusal rate is 0.0%** across every category, and the control/coherence prompts are clean. Throughput was **30.2 gen tok/s** mean (single-stream, full GPU offload) — roughly 3× faster per token than the dense Gemma-12B comparison, reflecting the ~3B-active MoE routing.
+> Caveats: the classifier is heuristic, not an LLM judge; the audit ran on the **Q4_K_M quant**, not the bf16 weights in this repo (representative, not identical). The StrongREJECT-based protocol in the Nous-Heretic method paper supersedes this heuristic pass.
 ## Intended use
 - **Security research and red-teaming** — probing model behavior without refusal confounds.